Mercurial > pylearn

annotate pylearn/datasets/MNIST.py @ 818:f4729745bb58

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
backporting to 2.4
author dumitru@deepnets.mtv.corp.google.com
date Wed, 02 Sep 2009 14:22:02 -0700
parents f30bb746f279
children bfc5914642ce
rev   line source
470
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
1 """
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
2 Various routines to load/access MNIST data.
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
3 """
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
4 from __future__ import absolute_import
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
5
505
74b3e65f5f24 added smallNorb dataset, switched to PYLEARN_DATA_ROOT
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 504
diff changeset
6 import os
470
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
7 import numpy
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
8
818
f4729745bb58 backporting to 2.4
dumitru@deepnets.mtv.corp.google.com
parents: 795
diff changeset
9 from pylearn.io.pmat import PMat
f4729745bb58 backporting to 2.4
dumitru@deepnets.mtv.corp.google.com
parents: 795
diff changeset
10 from pylearn.datasets.config import data_root # config
f4729745bb58 backporting to 2.4
dumitru@deepnets.mtv.corp.google.com
parents: 795
diff changeset
11 from pylearn.datasets.dataset import Dataset
470
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
12
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
13 def head(n=10, path=None):
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
14 """Load the first MNIST examples.
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
15
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
16 Returns two matrices: x, y. x has N rows of 784 columns. Each row of x represents the
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
17 28x28 grey-scale pixels in raster order. y is a vector of N integers. Each element y[i]
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
18 is the label of the i'th row of x.
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
19
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
20 """
818
f4729745bb58 backporting to 2.4
dumitru@deepnets.mtv.corp.google.com
parents: 795
diff changeset
21 if path is None:
f4729745bb58 backporting to 2.4
dumitru@deepnets.mtv.corp.google.com
parents: 795
diff changeset
22 path = os.path.join(data_root(), 'mnist','mnist_all.pmat')
470
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
23
794
951272679910 get the mnist data from the pmat file and not the amat file
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 658
diff changeset
24 dat = PMat(fname=path)
470
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
25
794
951272679910 get the mnist data from the pmat file and not the amat file
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 658
diff changeset
26 rows=dat.getRows(0,n)
504
19ab9ce916e3 slightly more sophisticated system for finding the mnist data
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 475
diff changeset
27
794
951272679910 get the mnist data from the pmat file and not the amat file
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 658
diff changeset
28 return rows[:,0:-1], numpy.asarray(rows[:,-1], dtype='int64')
470
bd937e845bbb new stuff: algorithms/logistic_regression, datasets/MNIST
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
29
795
f30bb746f279 put in comment a fct call all that seam to don't be used and conflict with numpy and python name.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 794
diff changeset
30
f30bb746f279 put in comment a fct call all that seam to don't be used and conflict with numpy and python name.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 794
diff changeset
31 #What is the purpose of this fct?
f30bb746f279 put in comment a fct call all that seam to don't be used and conflict with numpy and python name.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 794
diff changeset
32 #If still usefull, rename it as it conflict with the python an numpy nake all.
f30bb746f279 put in comment a fct call all that seam to don't be used and conflict with numpy and python name.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 794
diff changeset
33 #def all(path=None):
f30bb746f279 put in comment a fct call all that seam to don't be used and conflict with numpy and python name.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 794
diff changeset
34 # return head(n=None, path=path)
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
35
475
11e0357f06f4 typo in MNIST.train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 471
diff changeset
36 def train_valid_test(ntrain=50000, nvalid=10000, ntest=10000, path=None):
11e0357f06f4 typo in MNIST.train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 471
diff changeset
37 all_x, all_targ = head(ntrain+nvalid+ntest, path=path)
471
45b3eb429c15 added train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 470
diff changeset
38
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
39 rval = Dataset()
471
45b3eb429c15 added train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 470
diff changeset
40
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
41 rval.train = Dataset.Obj(x=all_x[0:ntrain],
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
42 y=all_targ[0:ntrain])
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
43 rval.valid = Dataset.Obj(x=all_x[ntrain:ntrain+nvalid],
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
44 y=all_targ[ntrain:ntrain+nvalid])
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
45 rval.test = Dataset.Obj(x=all_x[ntrain+nvalid:ntrain+nvalid+ntest],
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
46 y=all_targ[ntrain+nvalid:ntrain+nvalid+ntest])
471
45b3eb429c15 added train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 470
diff changeset
47
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
48 rval.n_classes = 10
563
16f91ca016b1 * added NStages as a stopper (moved from hpu/conv)
desjagui@atchoum.iro.umontreal.ca
parents: 537
diff changeset
49 rval.img_shape = (28,28)
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
50 return rval
475
11e0357f06f4 typo in MNIST.train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 471
diff changeset
51
11e0357f06f4 typo in MNIST.train_valid_test
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 471
diff changeset
52
627
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
53 def full():
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
54 return train_valid_test()
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
55
658
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
56 #usefull for test, keep it
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
57 def first_10():
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
58 return train_valid_test(ntrain=10, nvalid=10, ntest=10)
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
59
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
60 #usefull for test, keep it
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
61 def first_100():
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
62 return train_valid_test(ntrain=100, nvalid=100, ntest=100)
6d927441a38f added pylearn.datasets.MNIST.first_10 and pylearn.datasets.MNIST.first_100. They are usefull to test with small dataset.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 653
diff changeset
63
627
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
64 def first_1k():
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
65 return train_valid_test(ntrain=1000, nvalid=200, ntest=200)
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
66
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
67 def first_10k():
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
68 return train_valid_test(ntrain=10000, nvalid=2000, ntest=2000)
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
69
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
70 #old method from factory idea days... delete when ready -JB20090119
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
71 def mnist_factory(variant="", ntrain=None, nvalid=None, ntest=None):
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
72 if variant=="":
627
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
73 return full()
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
74 elif variant=="1k":
627
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
75 return first_1k()
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
76 elif variant=="10k":
627
ec27e19bb6eb moving away from mnist_factory
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 563
diff changeset
77 return first_10k()
537
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
78 elif variant=="custom":
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
79 return train_valid_test(ntrain=ntrain, nvalid=nvalid, ntest=ntest)
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
80 else:
b054271b2504 new file structure layout, factories, etc.
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 511
diff changeset
81 raise Exception('Unknown MNIST variant', variant)